Arctic Sea Ice Seasonal Prediction by a Linear Markov Model

Abstract

Abstract A linear Markov model has been developed to predict sea ice concentration (SIC) in the pan-Arctic region at intraseasonal to seasonal time scales, which represents an original effort to use a reduced-dimension statistical model in forecasting Arctic sea ice year-round. The model was built to capture covariabilities in the atmosphere–ocean–sea ice system defined by SIC, sea surface temperature, and surface air temperature. Multivariate empirical orthogonal functions of these variables served as building blocks of the model. A series of model experiments were carried out to determine the model’s dimension. The predictive skill of the model was evaluated by anomaly correlation and root-mean-square errors in a cross-validated fashion . On average, the model is superior to the predictions by anomaly persistence, damped anomaly persistence, and climatology. The model shows good skill in predicting SIC anomalies within the Arctic basin during summer and fall. Long-term trends partially contribute to the model skill. However, the model still beats the anomaly persistence for all targeted seasons after linear trends are removed. In winter and spring, the predictability is found only in the seasonal ice zone. The model has higher anomaly correlation in the Atlantic sector than in the Pacific sector. The model predicts well the interannual variability of sea ice extent (SIE) but underestimates its accelerated long-term decline, resulting in a systematic model bias. This model bias can be reduced by the constant or linear regression bias corrections, leading to an improved correlation skill of 0.92 by the regression bias correction for the 2-month-lead September SIE prediction.